« News headlines | Main | What is an unsession? »

Well-tie workflow

We've had a couple of emails recently about well ties. Ever since my days as a Landmark workflow consultant, I've thought the process of calibrating seismic data to well data was one of the rockiest parts of the interpretation workflow—and not just because of SynTool. One might almost call the variety of approaches an unsolved problem.

Tying wells usually involves forward modeling a synthetic seismogram from sonic and density logs, then matching that synthetic to the seismic reflection data, thus producing a relationship between the logs (measured in depth) and the seismic (measured in travel time). Problems arise for all sorts of reasons: the quality of the logs, the quality of the seismic, confusion about handling the shallow section, confusion about integrating checkshots, confusion about wavelets, and the usability of the software. Like much of the rest of interpretation, there is science and judgment in equal measure. 

← Synthetic seismogram (right) from the reservoir section of the giant bitumen field Surmont, northern Alberta. The reservoir is only about 450 m deep, and about 70 m thick. From Hall (2009), Calgary GeoConvention

I'd go so far as to say that I think tying wells robustly is one of the unsolved problems of subsurface geoscience. How else can we explain the fact that any reasonably mature exploration project has at least 17 time-depth curves per well, with names like JLS_2002_fstk01_edit_cks_R24Hz_final?

My top tips

First, read up. White & Simm (2003) in First Break 21 (10) is excellent. Rachel Newrick's essays in 52 Things are essential. Next, think about the seismic volume you are trying to tie to. Keep it to the nears if possible (don't use a full-angle stack unless it's all you have). Use a volume with less filtering if you have it (and you should be asking for it). And get your datums straight, especially if you are on land: make certain your seismic datum is correct. Ask people, look at SEGY headers, but don't be satisfied with one data point.

Once that stuff is ironed out:

  1. Chop any casing velocities or other non-data off the top of your log.
  2. Edit as gently and objectively as possible. Some of those spikes might be geology.
  3. Look at the bandwidth of your seismic and make an equivalent zero-phase wavelet.
  4. Don't extract a wavelet till you have a few good ties with a zero-phase wavelet, then extract from several wells and average. Extracting wavelets is a whole other post...
  5. Bulk shift the synthetic (e.g. by varying the replacement velocity) to make a good shallow event tie.
  6. Stretch (or, less commonly, squeeze) the bottom of the log to match the deepest event you can. 
  7. If possible, don't add any more tie points unless you really can't help yourself. Definitely no more than 5 tie points per well, and no closer than a couple of hundred milliseconds.
  8. Capture all the relevant data for every well as you go (screenshot, replacement velocity, cross-correlation coefficient, residual phase, apparent frequency content).
  9. Be careful with deviated wells; you might want to avoid tying the deviated section entirely and use verticals instead. If you go ahead, read your software's manual. Twice.
  10. Do not trust any checkshot data you find in your project — always go back to the original survey (they are almost always loaded incorrectly, mainly because the datums are really confusing).
  11. Get help before trying to load or interpret a VSP unless you really know what you are doing.

I could add some don'ts too...

  1. Don't just match the well picks to the seismic horizons. I have seen this go wrong lots of times, including places where 'everyone knows what the picks are'.
  2. Don't copy time-depth tables, well logs, or synthetics from one interval to another or one well to another — that way madness lies. If you need time-depth where you don't have it, the best idea is to build a velocity model.
  3. Don't use a different approach, wavelet, window, etc., for different intervals or different wells unless you have a really good reason (e.g. you're in different 2D vintages, or there are dramatic, mappable changes in lithology). 
  4. Don't tie wells to 2D seismic lines you have not balanced yet, unless you're doing it as part of the process of deciding how to balance the seismic. 
  5. Don't create multiple, undocumented, obscurely named copies or almost-copies of well logs and synthetics, unless you want your seismic interpretation project to look like every seismic interpretation project I've ever seen (you don't).

Well ties are one of those things that get in the way of 'real' (i.e. fun) interpretation so they sometimes get brushed aside, left till later, rushed, or otherwise glossed over. Resist at all costs. If you mess them up and don't find out till later, you will be very sad, but not as sad as your exploration manager.

PrintView Printer Friendly Version

EmailEmail Article to Friend

Reader Comments (7)

Hi Matt, great post!

You already know that I've been implementing the Guided Seismic-to-Well tie. As we mention in the abstract "Results compare well with manual seismic-to-well ties; yet we advise against unsupervised applications as the method is intended as a guide instead of a fully automated blind approach."

After a lot of work on this technique and now reading your post, I recognize that it is a hard task to emulate all the human perception and experience involved in the well-tie process. This is still an unsolved problem :). We don't have big pretentious with this method, just another tool that perhaps could help in one step of your workflow. Always used with common sense because the automatic tying can produce unphysical connections.

I am sharing the links of the abstract and the ppt, full paper coming soon.
Best regards,

abstract -
ppt -

April 25, 2013 | Unregistered CommenterHenry

@Henry: Thanks for the comment and awesome links! I am really interested in automated approaches — as you say, I think they can augment the manual workflow and perhaps help ensure that nothing unphysical is implied by the interpretation. It would be interesting to take approaches like yours and watch how interpreters want to apply them — can we do a better job of understanding where we need help, where the hardest decisions are, and what kind of solution is the most useful?

April 25, 2013 | Registered CommenterMatt Hall

Data Quality: A lot of it comes down to examining the quality and bandwidth of your seismic beforehand, and picking a good window.

Phase: On my current project, the phase of our wavelets vary (advance and lag) wildly between wells based on orientation of the geologic structure in comparison to acquisition orientation, amplitude focusing and image quality at the level of interest. All things to consider when throwing around goodness-of-fit numbers. I've seen HR offer up awesome cross-correlation numbers but the seismic and synthetic have nothing to do with one another.

Good stuff. Also just got done re-reading Simm and White (2003) before I pass it on to someone who is learning inversion.

April 25, 2013 | Unregistered CommenterMaitri

Thanks for the post!
My 2 cents: I also try to do more than one phase scan. I have seen many examples with major residual phase shift on land data. I usually do a first phase scan before any stretch and squeeze on a small window and then a second one after tieing top and bottom on a bigger window. Each time we force tie a point we loose a bit of phase information.
Hope it is useful.

April 25, 2013 | Unregistered CommenterFranck

I agree 99% but your point
6.Stretch (or, less commonly, squeeze) the bottom of the log to match the deepest event you can.

Please could you add to this something like....
Consider what any stretch/squeeze operation is telling you about your initial time-depth curve.

This is covered in Rachel Newrick's chapters but there is a danger of people reading your list and not taking the time to read her excellent words (even though READ THIS is prominently displayed!).

April 26, 2013 | Unregistered CommenterPeter

@Maitri: Interesting remarks about phase. I find 'phase' so mysterious, especially phase spectrums. I used to routinely get volumes from our processor in Calgary which were 90˚ out of phase every time (according to our well ties), despite (or because of?) their 'zero phasing' process. And I've seen people phase rotate their data by some very precise amount (23˚ say), only to decide later that they wished they hadn't... but you can't phase rotate all those horizons you picked.

I really like Liner (2002) and Roden & Sepulveda (1999) in this list of must-read papers.

April 26, 2013 | Registered CommenterMatt Hall

@Peter: Excellent point. In a way, it goes for everything in the process — indeed, every interpretive act implies something about the physical world. I've witnessed all sorts of unphysical, or at least highly improbable, universes implied by well ties: impossible wavelets, impossible velocities, impossible geology, etc. I do think software could do a better job of warning us about these scenarios when they crop up, but you're right, the interpreter needs to be vigilant.

I've put a wiki version of this post on SubSurfWiki. I added your point. Please feel free to edit or add!

April 26, 2013 | Registered CommenterMatt Hall

PostPost a New Comment

Enter your information below to add a new comment.
Author Email (optional):
Author URL (optional):
Some HTML allowed: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <code> <em> <i> <strike> <strong>